Domain Language Model Distillation for Independent Agent Deployment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing large language models (LLM) are inaccessible beyond their provided API, and there is a lack of methods for autonomous multi-agent frameworks that enable independent server deployment and collaboration with domain environments.
Innovation Solution
An autonomous agent system is developed to train a domain language model (DLM) by distilling knowledge from an LLM, utilizing a brain-mimicking approach where the hippocampus represents memory and the neo-cortex represents the LLM, allowing interaction with environments and other agents to enhance training through chain-of-thought prompting and self-consistency strategies.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If LLM is used through API only, then the model performance is maintained, but the system cannot be deployed independently on a server
Solution Approach 1:
The patent creates a student language model that copies and distills knowledge from the teacher LLM through iterative training processes. The student model learns reasoning patterns, chain-of-thought methodologies, and domain-specific knowledge from the LLM's API responses, enabling independent deployment while maintaining performance through knowledge transfer rather than direct model copying
Solution Approach 2:
The patent introduces an intermediary training system that facilitates knowledge transfer from the LLM to the student model. This intermediary process involves capturing LLM responses, extracting reasoning patterns, and systematically training the student model on these distilled knowledge, enabling independent server deployment without direct LLM dependency
2Adaptability or versatility
If distillation approach is used for PLM, then independent server deployment is enabled, but there is no method for autonomous multi-agent frameworks collaborating with domain environments
Solution Approach 1:
The patent develops a universal distillation framework that simultaneously enables independent deployment and autonomous agent collaboration. The student model is trained not only on basic responses but also on chain-of-thought reasoning patterns and domain environment interactions, allowing it to perform multiple functions including independent reasoning, agent collaboration, and domain-specific task execution
Solution Approach 2:
The patent implements preliminary training of the student model on distilled knowledge from the LLM before deploying autonomous agents. This preliminary action includes pre-training the student model on reasoning patterns, domain knowledge, and interaction protocols, so that when autonomous agents operate independently, they already possess the necessary capabilities without requiring real-time LLM assistance
Data Source
AI summary
The present invention relates to an autonomous agent system for training a domain language model (DLM) based on a large language model (LLM) and an operating method thereof. The present invention proposes an approach that can overcome the dependency of the LLM in a multi-agent environment through a language model distillation procedure. The present invention proposes an autonomous agent technology that automates the process of consolidating experiences based on a memory by using a self-consistency technique and a chain-of-thought (CoT) reasoning.